If you opened your rank tracker in late September or October and saw rankings fall off a cliff, you may not have lost anything at all. Several SEO tools now admit that Google has been feeding some of them altered, incomplete, or outright wrong search results for weeks. The tools call it “poisoning.” Google has said nothing about it, and the gap between what your tool reports and what real searchers see could be quietly distorting your client reports and your budgets.
Five vendors have now described how they detect suspect results, what they show you while a result is unconfirmed, and whether they fix numbers already stored. The answers differ a lot, and if you rely on rank data, you should know which kind of answer your tool gives.
What a “poisoned” result actually looks like
The pattern was first measured by rank-tracking company Nozzle. In an article published October 10 and updated October 21, Nozzle described result pages missing many, and sometimes all, of the sites that normally rank for a keyword. Nozzle counts a first page as poisoned when fewer than half of the sites that usually appear for a keyword show up anywhere in the collected results.
Measured that way, 8.7 percent of Nozzle’s first pages were poisoned between September 13 and October 21, against just 0.17 percent from August 20 to September 9, a jump of roughly fifty times. Nozzle’s own reading of the trend is blunt: Google is responding to AI scraping, and rank trackers are caught in the middle.
The practical consequence is simple. A page that looks like it dropped from position 4 to position 40 may never have moved in the results real people see. The tracker simply received a distorted page and recorded it as fact.
What each major tool says it checks
AccuRanker starts by throwing out shortened result pages, then fetches the page again in a fresh browser session. It compares the new page against the sites that ranked on page one for at least two of the previous three days, and only records a change when a second pull shows the same new lineup. One honest caveat: about one in twenty desktop pages and one in twelve mobile pages lack enough history for this check to run at all.
SEOmonitor takes a similar history-based approach. It estimates that around 8 percent of the results Google returns each day are incomplete or incorrect and get pulled again before being stored. On October 13 it switched to checking each result against the keyword’s own recent history as results arrive. Between October 9 and 13, though, many bad results stopped carrying the signals its checks depended on, and roughly 10 percent of monitored keywords were affected in that window.
Ahrefs says it runs filters that detect and reject abnormal results and request replacements, and it is now cleaning up past results that may have been affected. Its estimate is wide, 5 to 20 percent of search results affected over recent weeks depending on the data provider, and its statement does not say which dates, reports, or metrics the cleanup covers.
BrightLocal, focused on local search, is adding extra data suppliers so that if one provider fails, another can cover. Its CEO has been candid that no single fix solves this completely, for BrightLocal or anyone else. SerpApi, which sells Google results through an API, shipped an update on October 14 that it says should reduce the number of poisoned results returned, while noting work continues.
What happens to the numbers already stored
Historical data is where the damage lingers longest, because old exports do not update themselves.
SEOmonitor says it reprocessed keywords affected by bad September results from September 25 to 28, and that it blocked bad data from entering the product until October 8. For the October 9 to 13 window it reprocessed affected keywords on October 13, though it does not say which earlier daily values were revised, and it notes most movements in its Visibility metric were minimal.
AccuRanker runs a “sandwich check” over stored data, flagging days where fewer than half of the previous day’s page-one sites match the next day’s. On a sample of keywords, the flag rate was 30.2 percent on September 25 and just 0.58 percent on October 17. The company does not say whether flagged days get revised, and Ahrefs says it is cleaning up past results without detail on scope.
The lesson: an export you downloaded in late September may no longer match what the same tool shows today for the same dates. Before you resend old numbers to a client, compare the old export against the live tool and ask the vendor whether stored data for those dates was revised.
A five step check before your next report
You do not need to abandon rank tracking. You need a verification habit for the affected period, roughly mid-September through late October.
First, cross-check any dramatic drop in Google Search Console. It measures your own site’s performance from Google’s side, so if it shows no matching fall in clicks or impressions, treat the tracker drop as suspect.
Second, check your tool’s own incident or status page. SEOmonitor maintains one, Ahrefs says updates will appear in its dashboard, and AccuRanker’s blog documents its checks. If your vendor has published nothing, that silence is itself information.
Third, look at the shape of the drop. A genuine loss usually moves gradually or alongside a known change. A one-day cliff across many unrelated keywords, followed by a bounce back, smells like bad data, not a penalty.
Fourth, manually refresh the affected keywords where your tool allows it, and compare the fresh pull against the stored history.
Fifth, when a client asks about a September or October dip, say what you know and what you do not. Admitting the data is uncertain beats a confident story built on numbers you cannot defend.
Why this matters beyond one bad month
Rank data feeds budgets, staffing decisions, and client trust. If a tool reports a collapse that never happened, a team might rewrite pages that were fine, shift spend away from terms that were actually performing, or spend weeks chasing a penalty that does not exist. The reverse is just as risky: a real drop hidden inside a noisy period could go unnoticed because everyone assumes the data is broken.
The deeper issue is that none of these vendor statements cite independent testing of how well their checks work, and the figures they publish measure different things. You cannot line these numbers up side by side and declare a winner.
Until Google explains what changed, stay skeptical but practical: keep tracking, verify the big moves, and stop treating any single tool’s daily number as ground truth.

FAQ
What does “poisoned” search results mean?
It means a rank tracker received a results page missing many of the sites that normally rank for a keyword, so it may record a false drop. Nozzle counts a page as poisoned when fewer than half of the usual first-page sites appear anywhere in the collected results.
How widespread was the problem?
Nozzle measured 8.7 percent of first pages as poisoned between September 13 and October 21, up from 0.17 percent in late August to early September. Ahrefs estimated 5 to 20 percent of results affected depending on provider. SEOmonitor said about 10 percent of monitored keywords had incorrect results between October 9 and 13.
Did Google confirm it is altering results for trackers?
No. As of October 22, none of the vendor statements included a statement from Google, and AccuRanker explicitly noted that Google has made no statement. Nozzle’s view is that the trend is Google’s response to AI scraping.
Should I stop using rank trackers?
No. Rank data is still useful, but for the mid-September to late October period, treat dramatic moves as provisional until verified. Keep tracking, verify big changes against Search Console, and note in client reports when a figure comes from an affected window.